A 48-year-old woman presented with partial paralysis of the left hand due to peripheral nerve injury leading to chronic median and ulnar nerve neuropathy. She was involved in a multi-vehicle accident in 2019. Two days later, she noticed an impairment in her ability to move her left thumb and other fingers limiting her gripping and functional activities. It was assessed that the absence of left thumb movement appeared to be neurological in nature possibly originating from neck or thoracic outlet. She reported neck pain with left radicular symptoms, decreased range of movement in her cervical spine. An electrodiagnostic study conducted in July 2020 observed no evidence of cervical radiculopathy. Both studies repeatedly found median and ulnar nerve neuropathy at the wrist.

At the onset of this study, the participant had severe atrophy of the left abductor pollicis brevis (APB) muscle with no motor units. The median motor nerve was nonreactive at the wrist, and severe ulnar neuropathy at the elbow. The participant was enrolled into this current study in December 2020. The clinical assessment scores taken pre-intervention, at the end of intervention and at follow-up sessions are provided in Supplementary Table S1. All study procedures were approved by the Northwell Health Institutional Review Board. The study was conducted in accordance with the principles embodied in the Declaration of Helsinki and in accordance with local statutory requirements. The participant provided formal written informed consent to participate in this study. The study has been registered with ClinicalTrials.gov (NCT04755699, first posted on 16/02/2021).

Bipolar surface EMG electrodes were used to record muscle activity from the following muscles of the left arm and hand: biceps brachii (BIC), triceps brachii (TRI), flexor digitorum superficialis (FDS), extensor digitorum communis (EDC) and abductor pollicis brevis (APB). We chose these muscles as they are primarily innervated by the cervical roots, namely BIC: C5, TRI: C7, FDS: C8, EDC: C7, ABP: C7. We used pre-gelled Ag/AgCl electrodes (Ø 24 mm, Arbo™ H124SG, Covidien), connected to EMG sensors (AT-04-001 MyoWare™ Muscle Sensor), a differential amplifier (AD6221) with a 2nd order bandpass filter (20-498 Hz), and a signal digitizer (PicoScope® Model 4824A). The MyoWare sensor required the electrodes to have a center-to-center distance of 30 mm. The sampling rate used was 10 MHz while characterizing the recruitment profile of the upper limb muscles and 10–20 kHz during task performance. For characterizing the area-under-the curve (AUC), the EMG signal was digitally filtered using a 60 Hz IIR comb filter and a 6th-order Butterworth bandpass filter between 10–1,000 Hz in MATLAB.

Transcutaneous spinal cord stimulation was provided using a custom-built stimulator and electrode array. The stimulator consisted of a microcontroller to produce the stimulation waveform digitally, a Texas Instruments class-D amplifier (TAS5825P) in voltage mode, and a 12:1 step-up transformer (Xicon 42TM003-RC). The stimulator was powered by a 24 V battery and a safety circuitry was used to restrict power density levels to no more than 0.25 W/cm². The flexible PCB electrode array consisted of electroless nickel immersion gold (ENIG) or immersion silver-plated square contacts (10 mm x 10 mm) arranged in an 8 × 5 pattern with a 11 mm center-to-center electrode separation. Electrical stimulation to any combination of the 40 electrodes within the array, was controlled using solid-state relays mounted above each of the electrodes and a custom MATLAB-based GUI. Miniature green LEDs mounted on the dorsal surface of the electrode array, visually indicated the active electrodes. To target specific cervical segments (Figure 1A), an electrode configuration of 1 × 3 (Figure 1B) was used wherein stimulation was provided simultaneously to 3 adjacent electrodes within a single row, spanning the midline. The electrode array was affixed to the back of the neck using a rectangular piece of proprietary hydrogel. To ensure consistency in placement of the array between sessions, we used the inion of the external occipital protuberance as a landmark. Distances measured from the inion were used to place the electrode array and identify the location of stimulation. Two interconnected 5 × 10 cm rectangular self-adhesive hydrogel electrodes (Axelgaard Manufacturing Co., Ltd., United States) placed along the midline over the lumbar spinal cord served as return electrodes (anodes). Stimulation consisted of a 10 kHz biphasic sinusoidal waveform with pulse duration of 1 ms delivered at 3 Hz for generating recruitment profiles.

Stimulation frequency was set to 50 Hz when tSCS was paired with activity-based training and the pulse width was reduced to 0.5 ms to reduce neck muscle activation and increase participant comfort. Stimulation amplitude was gradually raised up to the maximal level that the participants found tolerable while not interfering with their training, which was usually 140–160 mA. tSCS paired with activity-based training was administered for 45–60 min per session.

Heart rate and blood pressure were measured at the beginning and end of each session. For finer measurements of voluntary muscle control, we measured both the evoked force and electromyographic (EMG) activity in the muscles of the affected hand and arm (left).

The tasks involved activating the D1 interphalangeal joint and separately generating maximal force while performing a lateral pinch grasp. For this, a custom rig was used. The participant was asked to push down on the sensor (load range of 10 lb., FX29 compact compression load cell, TE Connectivity) using their thumb. For all force measuring tasks, the experimenter cued the participants to push against the force sensor for 3–5 s and then to relax for 3–5 s. The task was performed in 2 sets of 5 trials each with a 60–90 s of rest period between the sets. Verbal encouragement was provided to encourage the participants to generate maximal force and to sustain (if possible) for the full cued period. The sensors used were calibrated at regular intervals using standard weights (Supplementary Figure S1).

While receiving tSCS, the participant performed the same tasks as described above. This served the dual purpose of providing activity-based training while receiving tSCS as well as measuring the volitional EMG activity and force.

Stimulation consisted of a 1 ms-long multiphasic pulse of the 10 kHz sinusoidal waveform at a stimulation frequency of 3 Hz. Stimulation amplitudes tested ranged from 100 mA to up to 225 mA in intervals of approximately 25 mA. This was repeated for each of the eight rows of electrodes on the array. The Picoscope 6 acquisition software (Pico Technology, Cambridgeshire, United Kingdom) was used to trigger acquisition of a 100 ms-long EMG signal following each stimulation pulse. For each stimulation amplitude and at each electrode row, we recorded such an EMG signal from 20–30 repetitions of the stimulation pulse. All the data was imported into MATLAB for further analysis. From each of the EMG signals recorded, we isolated a snippet starting from 5 ms and ending at 55 ms after the stimulation artifact. We measured the peak-to-peak amplitude (P2P) for this snippet. It was included for analysis only if the maximum amplitude of the snippet was greater than 5 times the standard deviation of baseline signal of that recording channel. For each muscle, the P2P amplitudes were normalized to the maximal P2P amplitude recorded across all amplitudes and electrode rows.

X-ray images for the participant in the sagittal plane with radio-opaque markers on the neck were obtained to determine actual location of electrodes with respect to spinal roots and vertebral landmarks.

We also measured the stimulation current of maximal activation for each muscle. For this, at each cervical level, we increased the amplitude until the reflexive activation of the muscle was visually determined to be consistent for every stimulation pulse and did not change for any small increase in amplitude. This stimulation amplitude was noted as the Maximal Activation Current for that muscle at the cervical level. This procedure was repeated for each muscle.

In addition to the areas recommended in the GRASSP assessment for sensibility evaluation, we tested additional relevant spots on the volar aspect of the hand to obtain a comprehensive profiling of the sensory changes in the entire hand. Specifically, we included the remaining finger pads and four corners of the palm. To provide encouragement and also characterize the true capability of the participant, we let the GRASSP prehension assessment tasks exceed the stipulated time of 75 s.

For statistical analysis of the evoked force, we compared the forces recorded in three stages of the study (start, no-stim and end of study). For the early stage, we chose the first three consecutive sessions after the start of weekly tSCS sessions. For the no-stim stage, we chose three sessions during the period when tSCS was not being delivered. For the end of study stage, we chose the last three consecutive sessions. For the force measurements involving the D1 IP muscle in case of the participant with the peripheral nerve injury, we performed a t-test. We employed the Bonferroni correction for all multiple comparisons.

We characterized the recruitment profile of the different motor pools innervating the upper limb muscles. To determine the effect of the location of stimulation along the rostro-caudal axis on the recruitment of the upper limb muscles, we sequentially delivered stimulation to different electrode triplets in the array (Figure 2A). Simultaneously, we recorded EMG activity from 5 muscles of the left arm and hand. We hypothesized that the recruitment pattern through transcutaneous stimulation would reflect the rostrocaudal segment-wise distribution of the upper limb motor nuclei in the cervical spinal cord. Almost all the muscles exhibited the best recruitment when stimulation was delivered around the C5-C6 level. Stimulating using rostral electrodes primarily activated the biceps (Figure 2B). Activation of the triceps invariably was strongest when stimulation was localized around the C5-C6 level. Stimulation using electrodes positioned more caudally resulted in decreased recruitment across all muscles. In participant CTS01, APB showed only weak recruitment irrespective of the location of stimulation, likely due to neuropathy. But we observed that APB was maximally recruited when stimulation was targeted at the C4-C5 level. Additionally, we also determined the current required for maximal activation (Max. Activation Current) for each of the upper limb muscles. The thresholds of activation showed similar results as the recruitment profiles with biceps and/or triceps muscles having a low threshold of activation during rostral stimulation (Figure 2C).

We focused on the volitional control of the D1 interphalangeal (D1 IP) joint and lateral pinch. We chose the site of stimulation based on the choice of muscle and its activation profile. For the participant CTS01, we focused on thumb muscles, and delivered tSCS targeted at the C5 level (about 8 cm from the inion of the external occipital protuberance) based on the activation profile for the APB muscle.

We observed an average increase of 407% (t-statistic = −5.26, p < 0.05) in the force generated by the flexion of D1 IP joint (Figure 3A) in the ‘No Stim’ period (7.3 ± 0.93 N) at the end of 16 sessions compared to the forces at the start of the study (1.79 ± 1.24 N). This reduced to a 287% increase (5.15 ± 1.22 N) by the end of the study (30 sessions). This was accompanied by an increase in EMG activity of the left APB muscle (Figure 3B). We did not observe a significant increase in the lateral pinch force during this period.

To evaluate the clinical significance of the progress showed by the participant, we also performed standard clinical assessments, namely the GRASSP test. Most of the improvement was observed in the thumb as well as the ulnar side of the hand. GRASSP strength showed increases in D1 IP and D5 abduction movements (Supplementary Figure S2). GRASSP sensibility showed up to a 3-point increase in sensation at the pinky tip (Supplementary Figure S2). In the nut threading task included in the GRASSP prehension tests, the participant showed an improvement in performing the correct grasp (Figure 4A) and a decrease in the time taken to perform the task (Figure 4B). The participant also showed improvements in the range of movement of thumb abduction, flexion, and extension as well as index abduction (Supplementary Figure S3). A 1-year follow-up EMG and nerve conduction evaluation was administered. While the study stated that the sensory nerves remained unchanged, the motor study of both the median and ulnar nerves showed improvements.